google-deepmind/alphafold3

Running on CUDA 12.2

Closed this issue · 1 comments

Hi DeepMind team,

Thank you for your time and support.

If I just modified (line number 11) the DockerFile to use cuda 12.2 instead of 12.6 and rebuild the docker. Would that be still a safe workaround? If not, what would be the consequences?

Thank you in advanced.

Please see [https://github.com//issues/18#issuecomment-2473001985](this comment):

We have only done full accuracy testing with CUDA 12.6, but we also tried CUDA 12.2 and AlphaFold 3 ran on that configuration as well -- but I can't give any guarantees about prediction accuracy with that configuration.

I strongly recommend using CUDA 12.6.